# Python Design Patterns > Python design pattern guidance. Use when choosing, implementing, or reviewing idiomatic Python patterns such as dataclasses, value objects, protocols, context managers, dependency injection, adapters, repositories, factories, CLI/service boundaries, async resource handling, pytest fixtures, or Clean/Hexagonal/Onion expression in Python. Use python-antipatterns for smell-focused audits and python-engineering for broader Python workflow. - Skill: `nledford/python-design-patterns` (Agent Skill) - Install (CLI): `npx skillmds@latest add nledford/python-design-patterns` - Raw SKILL.md: https://api.skillmd.com/api/skills/nledford/python-design-patterns/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Productivity - Author: nledford (https://skillmd.com/u/nledford) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/nledford/python-design-patterns --- # Python Design Patterns Use this skill to choose Python patterns that fit the codebase, runtime, and domain. Prefer simple modules, functions, data types, protocols, and explicit boundaries over pattern catalogs copied from class-heavy languages. ## Use When - Python code needs a domain model, adapter boundary, test seam, construction pattern, resource-lifetime pattern, or API structure. - A refactor should make behavior easier to test without real frameworks, databases, filesystems, or services. - Clean, Hexagonal, Onion, DDD, BDD, or TDD guidance needs Python-specific expression. - Review feedback asks for a more idiomatic Python design. For smell-focused review, use [`python-antipatterns`](../python-antipatterns/SKILL.md). For package, uv, typing, test, and command workflow, use [`python-engineering`](../python-engineering/SKILL.md). Advice-only pattern choice may use this skill independently. When applying or implementing a pattern, load [`python-engineering`](../python-engineering/SKILL.md). ## Selection Rules 1. Inspect local Python version, package layout, tests, framework conventions, typing policy, and existing data-model choices. 2. Prefer plain functions and cohesive modules until identity, lifecycle, invariants, substitution, or state justify classes or protocols. 3. Use structural typing and dependency injection at real boundaries, not for every collaborator. 4. Keep framework, ORM, request, response, serializer, and SDK types at adapters unless the project intentionally centers that framework. 5. Test domain rules with ordinary unit tests; test adapters with real framework, database, filesystem, or service contracts. ## Patterns To Prefer - **Dataclasses and frozen dataclasses:** use for explicit data shapes, value objects, and immutable domain values. Add validation constructors when invalid states matter. - **Enums and literal types:** use for small closed sets and states where string values would otherwise drift. - **Protocols:** use for outbound ports, plugin surfaces, or collaborators where structural substitution matters. Keep protocols small and behavior-oriented. - **Context managers:** use for resources that need reliable setup and teardown: files, locks, transactions, temporary directories, spans, and test fixtures. - **Factories and constructors:** use factory functions or classmethods when creation has validation, defaults, environment lookup, or dependency wiring. - **Adapter modules:** put HTTP clients, database access, queues, filesystems, CLIs, and framework handlers behind boundary modules that translate to domain concepts. - **Repository pattern:** use only when persistence shape should be hidden from domain/application policy. Do not create generic CRUD repositories by default. - **Functional core, imperative shell:** keep parsing, validation, decisions, and transformations pure where practical; keep I/O at the edges. - **Pytest fixtures/builders:** use fixtures for resource setup and builders for readable test data, with explicit cleanup and deterministic values. ## Async Ownership Patterns - Prefer a supervised task group when sibling tasks share a lifetime: the parent creates them, observes failures, propagates shutdown, and awaits every child before leaving its resource scope. - Give each queue, finite worker set, and the resources those workers use one explicit lifecycle owner. A task that must outlive a caller needs a named, longer-lived supervisor rather than becoming detached work. Route cancellation, backpressure, overload, retry, ordering, recovery, and test workflow to [`python-engineering`](../python-engineering/SKILL.md) and its [asyncio concurrency reference](../python-engineering/references/asyncio-concurrency.md). ## Architecture And Methods - **DDD:** use value objects, entities with identity, domain services, policies, and explicit state transitions where they protect real rules. - **Clean / Hexagonal / Onion:** represent use cases as functions or application service classes; use protocols for outbound ports when tests or adapters need substitution; keep framework handlers thin. - **BDD:** express externally observable workflows with domain language before choosing pytest, behave, pytest-bdd, or plain test names. - **TDD:** write focused tests for parsers, validators, policies, state changes, and error mapping before wiring broad integrations. ## Review Checklist - Does the pattern solve a current behavior, boundary, or testing problem? - Are type hints meaningful without replacing runtime validation at trust boundaries? - Are exceptions specific and preserved with `raise ... from exc` where useful? - Are resources closed deterministically through context managers, fixtures, or framework lifecycle hooks? - Are framework and persistence models kept out of core domain APIs where that boundary matters? - Are mocks limited to external, slow, nondeterministic, or infrastructure boundaries? ## Common Mistakes - Adding abstract base classes or protocols for one concrete implementation. - Turning simple functions into manager/service classes with no state or boundary. - Treating Pydantic, ORM, or serializer models as domain objects by default. - Hiding environment, filesystem, network, or database access in import-time side effects. - Using fixtures or monkeypatching to replace the behavior being specified.